Building a multi-camera vision application on Orin NX 16GB and would like to confirm decode capacity before finalising our hardware.
Our load: 4 concurrent RTSP streams, H.265 Main profile, 2560×1440 (4MP), 15–20 fps each, decoded via DeepStream and fed to a detection model.
The module datasheet lists decode as 2× 4K60, 4× 4K30, 9× 1080p60, 18× 1080p30 (H.265).
Three questions:
Our 4MP streams sit between 1080p and 4K. Is it correct to treat 4× 2560×1440 at 20fps as comfortably within the 4× 4K30 figure, or is there a pixel-rate limit I should calculate against instead?
Is NVDEC capacity affected by the power mode? Specifically, is decode throughput the same at 25W as at the 40W MAXN SUPER profile, or does it scale with clocks?
What is the correct way to verify at runtime that decoding is genuinely running on NVDEC rather than silently falling back to software? We want to confirm this on the bench rather than assume it.